feat(cli): add runtime completers for sandbox/cluster/provider names #41

Activity

1. pimlock

   added

   enhancement

   state:agent-readyApproved for agent implementationApproved for agent implementation

   mr-opened

   on Feb 27, 2026
2. pimlock commented on Feb 27, 2026

   

   pimlock

   on Feb 27, 2026

   CollaboratorAuthor

   More actions

   🏗️ build-plan

   Implementation Plan

   Issue type: feat
   Complexity: Medium
   Confidence: High — clear path, two critical blockers already identified with solutions

   Summary

   Add runtime tab-completion for sandbox, cluster, and provider names by creating completer functions that fetch live data from the Navigator gateway (or local filesystem for clusters). This requires a new completers.rs module with shared helpers, wiring ArgValueCompleter onto 10 existing positional arguments, and fixing two initialization-order issues in main.rs.

   Scope

   • crates/navigator-cli/src/completers.rs (NEW, ~130 lines): Three public completer functions (complete_sandbox_names, complete_cluster_names, complete_provider_names) plus shared helpers (blocking_complete for thread-spawning async gRPC calls, completion_grpc_client with 2s timeout)
   • crates/navigator-cli/src/lib.rs (1 line): Add pub mod completers;
   • crates/navigator-cli/src/main.rs (~30 lines modified): Move crypto provider init before CompleteEnv::complete(), add #[arg(add = ArgValueCompleter::new(...))] to 10 target arguments, add import

   Critical Blockers (Already Solved)

   1. Crypto provider init order — rustls::crypto::aws_lc_rs::default_provider().install_default() currently runs AFTER CompleteEnv::complete(). Any TLS-based completer would fail. Fix: Move init before complete().

   2. Nested tokio runtime — #[tokio::main] creates a runtime, and complete() runs synchronously inside block_on. Creating another runtime on the same thread panics. Fix: Spawn a dedicated OS thread via std::thread::spawn + tokio::runtime::Builder::new_current_thread() for async gRPC calls.

   Target Arguments (10 total)

   Command	Field	Completer
   SandboxCommands::Get	name: String	complete_sandbox_names
   SandboxCommands::Delete	names: Vec<String>	complete_sandbox_names
   SandboxCommands::Connect	name: String	complete_sandbox_names
   SandboxCommands::Sync	name: String	complete_sandbox_names
   ForwardCommands::Start	name: String	complete_sandbox_names
   ForwardCommands::Stop	name: String	complete_sandbox_names
   ClusterCommands::Use	name: String	complete_cluster_names
   ProviderCommands::Get	name: String	complete_provider_names
   ProviderCommands::Update	name: String	complete_provider_names
   ProviderCommands::Delete	names: Vec<String>	complete_provider_names

   Implementation Steps

   1. Move crypto provider init before CompleteEnv::complete() in main.rs — Zero-risk reorder. The init is idempotent and has no dependencies on the tokio runtime.

   2. Create crates/navigator-cli/src/completers.rs with:

      • complete_cluster_names(prefix: &OsStr) -> Vec<CompletionCandidate> — Uses list_clusters() from navigator-bootstrap (local filesystem scan, no network call, fast)
      • blocking_complete(future) -> Vec<CompletionCandidate> — Shared helper that spawns an OS thread, creates a current_thread tokio runtime, runs the async future, returns results (or empty vec on any error)
      • completion_grpc_client(endpoint) -> NavigatorClient — Async helper that builds a gRPC client with 2s connect timeout and mTLS (reuses require_tls_materials + build_tonic_tls_config from tls.rs)
      • complete_sandbox_names(prefix: &OsStr) -> Vec<CompletionCandidate> — Resolves active cluster, calls blocking_complete with gRPC ListSandboxes
      • complete_provider_names(prefix: &OsStr) -> Vec<CompletionCandidate> — Resolves active cluster, calls blocking_complete with gRPC ListProviders
      • All completers return empty vec on any error (unreachable cluster, no active cluster, TLS failure) — never block the shell

   3. Add pub mod completers; to lib.rs

   4. Wire #[arg(add = ArgValueCompleter::new(...))] on all 10 target arguments in main.rs

   5. Add tests:

      • Unit tests in completers.rs: verify complete_cluster_names returns results when metadata files exist, returns empty vec when config dir is missing
      • Graceful-failure test: verify sandbox/provider completers return empty vec when no active cluster is configured (no panic, no hang)

   6. Run mise run pre-commit and verify all checks pass

   Test Plan

   • Unit tests: In completers.rs — test cluster completer with mock filesystem, test sandbox/provider completers return empty vec gracefully when cluster is unavailable
   • Integration tests: N/A for initial PR — gRPC mock server testing deferred to avoid scope creep. The completer's error-resilience is validated by unit tests.
   • E2E tests: N/A — completions are interactive shell features; manual smoke testing with navigator sandbox get <tab> against a live cluster

   Risks & Open Questions

   • Shell hang risk (Medium): Mitigated by 2s timeout on the gRPC client. If a cluster is unreachable, completion returns empty in ≤2s.
   • Large resource counts (Low): Completers don't paginate; capped at whatever the first response returns. Acceptable for completion UX (unlikely to have >200 sandboxes/providers).
   • --cluster flag unavailable to completers (Low): Clap hasn't parsed args when completers run, so completers use NAVIGATOR_CLUSTER env var or active_cluster file. This matches the common case.
   • Inference route completions: Deferred to a follow-up issue to keep this PR focused on the three core resource types mentioned in the issue.

   Documentation Impact

   • architecture/cli.md: Update with runtime completion architecture (completer functions, blocking bridge pattern, TLS init order)

   ---

   Revision 1 — initial plan

   ---

   Originally by @pimlock on 2026-02-26T19:39:43.699-08:00
3. pimlock commented on Feb 27, 2026

   

   pimlock

   on Feb 27, 2026

   CollaboratorAuthor

   More actions

   🏗️ build-from-issue-agent

   Will this show some kind of indicator when the names are being retrieved? Also it should fail gracefully — if something is not working (cannot connect, etc.) it shouldn't show anything.

   Loading indicator: No — there's no way to show a spinner or loading indicator during tab-completion. The shell blocks while waiting for the completer function to return, and the shell itself controls the UX. This is standard behavior for all CLI tools with dynamic completions (e.g., kubectl, docker, gh). The 2-second timeout ensures the shell doesn't hang long if the cluster is unreachable.

   Graceful failure: Yes — this is already covered in the plan. All completers return an empty Vec<CompletionCandidate> on any error (unreachable cluster, no active cluster configured, TLS failure, timeout, etc.). The user just sees no completions appear, same as if there were no matching resources. No error messages, no panics, no shell disruption.

   No changes to the plan needed — both behaviors are already accounted for.

   ---

   Originally by @pimlock on 2026-02-26T19:47:03.563-08:00
4. pimlock commented on Feb 27, 2026

   

   pimlock

   on Feb 27, 2026

   CollaboratorAuthor

   More actions

   🏗️ build-from-issue-agent

   Implementation Complete

   MR: !63

   What was built

   Runtime tab-completion for sandbox, cluster, and provider name arguments across 10 CLI commands. Cluster completions use local metadata (no network), sandbox/provider completions query the gateway via gRPC with a 2s timeout. All completers fail silently.

   Tests

   • Unit: 3 tests added (graceful failure for each completer type)
   • Integration: N/A
   • E2E: N/A

   Docs updated

   • None needed (no architecture/cli.md exists; inline comments document the approach)

   The issue will auto-close when the MR is merged.

   ---

   Originally by @pimlock on 2026-02-26T20:04:57.428-08:00
5. pimlock

   closed this as completedin #44on Feb 27, 2026
6. factory-octavian

   added a commit that references this issue on Mar 31, 2026

   chore: update readme (NVIDIA#41)

   e705056
7. Ygnas

   added a commit that references this issue on Sep 18, 2026

   Merge pull request NVIDIA#41 from opendatahub-io/konflux/mintmaker/ma…

   ...

   72d6b51